multixoR 0.2.1
Clarify the default MCTS score, half-credit draw convention, shipped heuristic calibrator and limits of its decisive-outcome training fit. Calibration diagnostics describe fitting data, not held-out performance.
Reject missing, fractional, nonnumeric and out-of-range player values in
mxo_evaluate()andmxo_win_prob()before coercion or terminal-state returns. Valid player perspectives and numerical results are unchanged.
multixoR 0.2.0
CRAN preparation
- Maintainer address changed to raban.heller@uni-ulm.de.
-
.Rbuildignorecovers the development folders, so the source package no longer contains an emptytools/directory; README links to the rules specification on GitHub. - Added spelling
WORDLISTentries and updatedcran-comments.md. - The three slowest tests (opening tables and an isotonic calibration fit, about two minutes together) are skipped on CRAN.
Bug fixes and improvements
Make human play functional in the Shiny controls, honor initial difficulty, add AI replies and explicit AI stepping, and check the required plotly dependency.
Forward named AI/rating engine arguments, including explicit
NULLbudgets.Bound hard AI with cooperative deadline checks through tree traversal and rollouts; use random rollouts for MCTS move ratings by default.
Rank search ratings by search value and leave unsupported probability labels missing. Reject invalid or fractional search depths before conversion.
Store distinct isotonic knots and handle constant-score calibration data.
multixoR 0.1.0
Initial release.
Engine
- Generic
n^dlattice geometry with 121 canonical directions for the default 4 / 3 / 3 configuration;(n, d_spatial, k)are first-class parameters of every engine function. -
mxo_gameS3 withmxo_new_game(),mxo_move(),mxo_play(),mxo_legal_moves(),mxo_undo(),mxo_replay(),mxo_status(),mxo_is_terminal(), plus type-stable accessors (mxo_board,mxo_timelines,mxo_history,mxo_to_move,mxo_config) andas_tibble.mxo_game. - Placement-anchored win detection across all five axes (spatial, time, timeline, mixed); branching never mutates the source timeline.
- Canonical ply notation via
mxo_format_ply()/mxo_parse_ply().
AI and evaluation
-
mxo_evaluate()— heuristic weighted m-line score with a cross-timeline multiplier (w_timeline). -
mxo_search()— negamax + α-β withbranch_policy(all/promising/none). -
mxo_mcts()— light UCT with random / heuristic rollouts,time_budget, seedable. -
mxo_win_prob()— heuristic logistic via a fitted internal calibrator (see Calibration below) or MCTS visit aggregation. -
mxo_rate_moves()— type-stable per-move tibble with rank + blunder/strong/best label. -
mxo_ai_move()— easy / medium / hard difficulty knob.
Simulation and strategy
-
mxo_policy()thin wrapper around the searchers;mxo_self_play(),mxo_simulate()with reproducible per-game sub-seeds. -
mxo_opening_table(),mxo_policy_tournament(),mxo_branch_study(),mxo_timeline_win_rate()(the §12 cross-timeline-win stress test). -
mxo_make_calibration_data()+mxo_fit_calibration()produce the default calibrator shipped inR/sysdata.rda;mxo_win_prob()resolves it via.mxo_active_calibrator().
Visualisation
-
mxo_plot_board()— Z-slices (ggplot2) and rotatable cube (plotly) toggles, withtop3andheatmapoverlays consuming amxo_rate_moves()tibble. -
mxo_plot_multiverse(),mxo_plot_threats(),mxo_plot_win_prob(),mxo_plot_eval(),mxo_plot_opening(),mxo_plot_tree(). -
autoplot()methods formxo_game,mxo_sim_result,mxo_game_record.
App and docs
-
mxo_run_app()launches the bundled Shiny app (shiny,bslib,DTin Suggests) with the multiverse / board / threats / tree / analysis tabs. -
mxo_example_game()deterministic short game with a branch, used by docs / vignettes / the app default state. - Four vignettes: Getting started, Rules and geometry, Analysis, Self-play and simulation.
- pkgdown site at https://r-heller.github.io/multixoR/ themed via the CTTIR suite’s
themakRtemplate. - GitHub Actions: R-CMD-check matrix (macOS / Windows / Ubuntu × release + devel + oldrel-1), pkgdown deploy, codecov.